Tuesday, August 1st 2023
A Closer Look at NVIDIA's Cinder Block Sized Air Cooler for RTX 4090 Ti—22 Heatpipes
Amidst reports that NVIDIA has shelved plans to release a new flagship RTX 40-series graphics card to top the current RTX 4090, we are getting even more pictures of the company's ingenious product design for what could have been the GeForce RTX 4090 Ti, or RTX TITAN (Ada). Pictures of the card's cooling solution has been in the news since January 2023, with more images surfacing in June. We are getting our first images of the cooler disassembled, revealing a startling look at the engineering effort NVIDIA put into this thing.
As we mentioned earlier, the "RTX 4090 Ti" features a unique ruler-shaped PCB that's oriented along the plane of the motherboard, rather than perpendicular to it, like any other add-on card. This is to minimize the spatial footprint of the PCB, and maximize volume for the cooler—which is 4 slots, thick, with its entire thickness dedicated to heat-dissipation, and no obstruction posed by the PCB.
Disassembling the cooler reveals that it has a large, continuous aluminium fin-stack spanning the entire length of the card. These fins are along the plane of the motherboard, just like the card's PCB. A jaw-dropping 22 heatpipes pull heat from the base plate that makes contact with all hot components on the PCB, including the GPU, VRM, and memory; spreading heat along the fin-stack.
The base-plate (from an older article), is a large continuous block of nickel-plated copper, with mirror finish over the contact-points where the card's "AD102" GPU, twelve GDDR6X memory chips, and numerous DrMOS components make contact with the card. It is from here that the 22 heatpipes pull heat.
There are a total of three fans—one of them is an intake located toward the obverse side of the card, which draws in fresh air; one of them is toward the tail-end, pulling hot air through the fin-stack, and exhausting through the reverse side of the card; the third one is a conveyor fan, located bang in the middle of the fin-stack, its airflow intersects both the other fans. The airflow from all three fans goes right through the fin-stack, unobstructed by the PCB along the way.
The RTX 4090 Ti has three breakout components besides the PCB. The first of these is a PCI-Express 4.0 x16 finger that's perpendicular to the PCB, so the card slots into your motherboard like just another add-on card. The second component is a satellite PCB located along the top of the card, with its 16-pin 12VHPWR power connector.
The third component is an innovative new cable that connects the power receptacle PCB with the card's main PCB. This is a series of two flattened non-braided copper cables conveying the 12 V DC power from the connector to the PCB in a 2-pin format. Each of the two cables is capable of 600 W continuous delivery, along with excursions within the ATX 3.0 specification. Besides the two power cables, you have the four signal pins from the 12VHPWR being conveyed to the PCB using a thinner set of cables.
NVIDIA went through all this trouble to create a cooling solution capable of taming a maxed out "AD102" GPU with all its 72 TPCs (144 SM) or 18,432 CUDA cores enabled. The card would find its place in NVIDIA's product stack had the Radeon RX 7900 XTX posed a threat to the RTX 4090 (it doesn't). It remains to be seen by how much AMD pushes up the performance of the already maxed out "Navi 31" GPU in the upcoming Radeon RX 7950 XTX, so NVIDIA could find a reason to respond with this card.
Source:
harukaze5719 (Twitter)
As we mentioned earlier, the "RTX 4090 Ti" features a unique ruler-shaped PCB that's oriented along the plane of the motherboard, rather than perpendicular to it, like any other add-on card. This is to minimize the spatial footprint of the PCB, and maximize volume for the cooler—which is 4 slots, thick, with its entire thickness dedicated to heat-dissipation, and no obstruction posed by the PCB.
There are a total of three fans—one of them is an intake located toward the obverse side of the card, which draws in fresh air; one of them is toward the tail-end, pulling hot air through the fin-stack, and exhausting through the reverse side of the card; the third one is a conveyor fan, located bang in the middle of the fin-stack, its airflow intersects both the other fans. The airflow from all three fans goes right through the fin-stack, unobstructed by the PCB along the way.
The RTX 4090 Ti has three breakout components besides the PCB. The first of these is a PCI-Express 4.0 x16 finger that's perpendicular to the PCB, so the card slots into your motherboard like just another add-on card. The second component is a satellite PCB located along the top of the card, with its 16-pin 12VHPWR power connector.
The third component is an innovative new cable that connects the power receptacle PCB with the card's main PCB. This is a series of two flattened non-braided copper cables conveying the 12 V DC power from the connector to the PCB in a 2-pin format. Each of the two cables is capable of 600 W continuous delivery, along with excursions within the ATX 3.0 specification. Besides the two power cables, you have the four signal pins from the 12VHPWR being conveyed to the PCB using a thinner set of cables.
NVIDIA went through all this trouble to create a cooling solution capable of taming a maxed out "AD102" GPU with all its 72 TPCs (144 SM) or 18,432 CUDA cores enabled. The card would find its place in NVIDIA's product stack had the Radeon RX 7900 XTX posed a threat to the RTX 4090 (it doesn't). It remains to be seen by how much AMD pushes up the performance of the already maxed out "Navi 31" GPU in the upcoming Radeon RX 7950 XTX, so NVIDIA could find a reason to respond with this card.
83 Comments on A Closer Look at NVIDIA's Cinder Block Sized Air Cooler for RTX 4090 Ti—22 Heatpipes
I wonder if the ATX format is not coming to an end ...
Given the size of the GC and CPU cooler ...
is looks sturdy like that and in case it melts on the other end just replace it easily and avoid the hardware fix soldering services.
I still don't understand why NVIDIA isn't slowly switching to AiO as the standard for the high-end segment. Other manufacturers like ASUS and Co. show how you can keep these performance monsters in check without having a hairdryer in your computer.
Besides, at 600w, you'd still need a huge radiator/pump. Maybe acceptable for custom builds, nightmare for self-contained products.
It just works, and it works flawlessly. No hassle to install or maintain, just plug and play.
More hard to install, more wires, more places to fail. Change paste? Forget it. Fans Fail? Forget it.
AIO or pure water is the way to go for powers over 500W.
Watercooling requires you to install the card AND the radiator, that is one more action so it cant be harder the other way around.
More wires? what? its at best the same amount of wires, aka the power connectors and at worst the radiator fans need to get their power elsewhere.
More places the fail? what? how? with watercooling the pump can fail, the tubing and tube connectors can fail and the fans can fail.
With a heatsink ONLY the fans can fail....
Change paste is no harder then with a waterblock and replacing fans...ok MAYBE you finally have a point there but usually its easy to find replacement fans.
Installing a card with mounted hsf is just the matter of installing the card itself. Installing a card with water cooler adds hassle of fitting the radiator.
"More wires."
A typical AIO has fans and pumps, that's two sets of wires. An aircooler has one set.
"More places to fail."
Again, fans only, vs fans and a pump.
"Change paste."
A waterblock does need paste too...
Assuming the rare occasion one needs to replace it, a graphics cards' HSF can be easily dismounted. Just take the cards out and grab a screwdriver. Have fun taking out an entire AIO before you get to reach those screws tho.
"Fans fail."
About the only valid case against air HSFs here. But still offset by the mere fact that an AIO has to deal with risks of pump failure and leaks, loose/broken fittings, etc.
Both can be made user replacable too, we've seen Gainward coolers with replacable fans.
There is no question whatsoever air is more fool proof, its not even a debate lol
You you split up your personal opinions, phobias etc with real engineering analysis of this abomination.
The heat sink is bigger, but it's still mounted to a separate PCB, tightened with screws, and fitted as a whole to the chassis.
The scale issues would also apply to a hypothetical AIO version.Dunno. Plenty of cases, even mid-towers, can accommodate 4 slots. How many can fit - say - 480mm radiators?
The point is almost no one will put one of these in some cramped case with literally no room whatsoever for a typical radiator. Even really tiny SFF cases can fit 240 radiators usually but they can't fit 4 slot cards. Space is most certainly not an issue.I mean we've seen pictures of this thing for like half a year or something at this point ? Who knows if it will even be released.
But really, this argument is silly because in 5 years' time nobody is going to be using that GPU because it will be so horribly obsolete.
I do believe though that all AIOs should allow for draining and refilling by end users, which would render the permeation issue moot. I'd like it even better if GPU and CPU mountings were identical so that you could use an AIO in either application. NZXT had something like this but of course it only worked with their AIOs.